Stable Cascade vs Stable Diffusion XL

[@Pixovert]

Stable Cascade Course: - cutt.ly/xwBnvtjV Stable Cascade Introduced: - bit.ly/StableCascadeNews Stable Diffusion XL: - bit.ly/SDXLintroduced SDXL Portals and Big Cats: - cutt.ly/2wVY8uAN Hardware with 20GB+ VRAM - bit.ly/3urmjqO Beginner's Guides KZbin - bit.ly/BeginCOMFYUISDXL Udemy - bit.ly/SDGenStart Approximate Timestamps (AI Generated): 00:00: Introduction to Stable Cascade and comparison with Stable Diffusion XL. 01:48: Overview of VRAM requirements and file structure for Stable Cascade. 03:14: Exploring results with Stable Cascade 04:32: Showcasing text rendering and artistic capabilities in Stable Cascade. 05:50: Detailed analysis of specific prompts and outcomes in Stable Cascade. 08:30: Comparative reflection on strengths and weaknesses of Stable Cascade. 10:00: Concluding thoughts on the distinct features of Stable Cascade. Join 🏆 this channel to get access to exclusive content and perks: bit.ly/XOVjoin

[@LouisGedo]

👋

[@Falkonar]

We can not run it inside ComfyUI with inpaint and esle yet?

[@mycelia_ow]

apparently not, it's only unofficially released.

[@axwellsonney]

We can now,i installed it today on comfy and it works very well and fast 👍🔥

[@JayS.-mm3qr]

What's better, ComfyUI, or Pinokio, for using Stable Cascade? 20gb of VRAM? not many have that. I heard they fixed the memory issue, yes?

[@anushdsouza9632]

What is the likliness of me running it with 4gb vram

[@FusionDeveloper]

(outdated reply, no longer accurate) if you get it to run in ComfyUI, it will probably take you 5-5000 hours to generate 1 image and require you using the fallback to CPU RAM. on 11GB VRAM it takes me 30-60 minutes. Not worth it for me. SDXL UltraSpice is taking me 4-70 seconds (depending on resolution/settings) so i'm sticking with it instead. For you, maybe try SD Turbo custom models if you can't run SDXL models with -lowvram. One of the big issues with StableCascade is that it currently doesn't use FP16, which speeds things up a lot.

[@sigmaclub2924]

Bruh, you can run it, go check it and also yesterday stability released checkpoint for comfy

[@FusionDeveloper]

A lot of progress has been made over the last 5 days. there now several Cascade models and several ways to process the models to create images and they only take about a minute.

[@tradehut2782]

I tried cascade yesterday. Prompt following is better than SDXL. Throwing random words in the prompt will hurt. Nowhere as good as Dall-E though. Image quality is better on current sdxl based models. Base model cascade > base model sdxl

[@creatycohouse3]

its good news. imaging many ppl rushing to finetuning cascade.

[@fontenbleau]

i thought Nvidia released a hack to use system RAM in addition to VRAM in latest driver. What about two cards? I bought new 4070 Ti Super + 2070 Super in total it's kinda 4090 by VRAM and Cuda cores, bus same 256 bit, i have 2 PSUs to power.

[@WallyMahar]

Can u 🫂 links if you talk about them😐

[@CleoCat75]

Why is Cascade so much slower?? SDXL with the Foooocus GUI can generate images in about 15 seconds with 60 steps. Stable Cascade however takes 3-4 minutes! What's going on? I verified Cascade IS using my GPU at 100% usage.... but it's just soooooo slow, wonder if I'm doing something wrong? (I have a 3080ti 12GB card)

[@flareonspotify]

i was wondering why my 16 gb of ram wasnt working

[@mycelia_ow]

I have 24GB of vram, I should be fine I think.

[@Trapped39]

Ideogram ai seeing stability AI open source 😂😂😂😂, why our same model type but open source available on internet

[@hatuey6326]

20 gigas vram !!!!! bon ben non alors !

[@RikkTheGaijin]

useless.. it's much MUCH worse than SDXL and it requires 20GB of Vram. Dead on arrival.

[@metronome003]

Lol I compared it to sdxl base model, it's definitely better, it's better at generating high resolution images than sdxl (doesn't hulucinate or duplicate as much) but at 1024p its lacking in the details department, try generating at a higher res as 1024p

[@MarcSpctr]

poor fella, wants everything for free, atleast go buy a good GPU if you are getting the software for free, atleast invest in hardware. Also CASCADE is in preview right now, and also is way way way better in terms of sticking to the prompt like Dall-E. Quality and stuff can be added later on using fine-tuned models, that is not what the priority of these models are. Go earn some money and buy a good gpu edit: or just wait for someone to feel bad for poor people like you and make it run on 4gb VRAM GPUs 😂

[@SkyyySi]

​@@MarcSpctrIt might come as a surprise, but a lot of people don't have +$2000 just lying around. And for a lot of people, that simply out of reach.

[@kademo1594]

⁠its a preview rn and some stability staff said on twitter that they are confident that they can half the vram usage

[@CoolhandLukeSkywalkr]

SO i can use it with my rtx 3090 cool ​@@metronome003